Update app.py

app.py

@@ -4,11 +4,13 @@ import torch
 from transformers import BertTokenizer, AutoModel
 import tensorflow as tf
 
+# Load tokenizer and BERT model for embeddings extraction
 model_name = "aubmindlab/bert-base-arabertv02"
 tokenizer = BertTokenizer.from_pretrained(model_name)
 bert_model = AutoModel.from_pretrained(model_name)
 bert_model.eval()
 
+# Load your trained RNN model
 model = tf.keras.models.load_model("rnn_Bi.h5")
 print("✅ Model loaded successfully!")
 
@@ -16,59 +18,88 @@ def get_bert_embedding(text, max_length=100):
     inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=max_length)
     with torch.no_grad():
         outputs = bert_model(**inputs)
+    # Use CLS token embedding as sentence embedding
     embedding = outputs.last_hidden_state[:, 0, :].numpy()
-    embedding = embedding.reshape(1, 1, 768)
+    embedding = embedding.reshape(1, 1, 768)  # shape (1, 1, 768)
     return embedding
 
 def predict_sentiment(text):
     embedding = get_bert_embedding(text)
     pred = model.predict(embedding)[0][0]
-    label = "إيجابي" if pred > 0.5 else "سلبي"
+    label = "Positive" if pred > 0.5 else "Negative"
     confidence = pred if pred > 0.5 else 1 - pred
-    return label, confidence
+    return label, f"Confidence: {confidence:.2f}"
 
-css = """
+# Custom CSS for soft and pleasant colors like GPT style
+custom_css = """
 body {
     font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
     direction: rtl;
     text-align: right;
+    background-color: #f5f7fa;
+    color: #202123;
 }
 .gradio-container {
     max-width: 600px;
-    margin: auto;
-    background: #f9f9f9;
-    padding: 20px;
-    border-radius: 12px;
-    box-shadow: 0 6px 15px rgb(0 0 0 / 0.1);
+    margin: 40px auto;
+    background: #ffffff;
+    padding: 25px 35px;
+    border-radius: 16px;
+    box-shadow: 0 8px 24px rgba(32, 33, 35, 0.1);
 }
 .gr-button {
-    background-color: #007bff !important;
+    background-color: #4caf50 !important; /* soft green */
     color: white !important;
-    font-weight: bold;
-    border-radius: 8px !important;
-    padding: 10px 25px !important;
-    font-size: 16px !important;
+    font-weight: 600;
+    border-radius: 12px !important;
+    padding: 12px 30px !important;
+    font-size: 18px !important;
+    transition: background-color 0.3s ease;
+}
+.gr-button:hover {
+    background-color: #45a049 !important;
 }
 .gr-textbox textarea {
     font-size: 18px !important;
-    padding: 10px !important;
+    padding: 14px !important;
+    border: 1.5px solid #d1d9e6 !important;
+    border-radius: 12px !important;
+    background-color: #f9fbfd !important;
+    color: #202123 !important;
+    transition: border-color 0.3s ease;
+}
+.gr-textbox textarea:focus {
+    border-color: #4caf50 !important;
+    outline: none;
+}
+.gr-label {
+    font-size: 20px !important;
+    font-weight: 600 !important;
+    margin-bottom: 8px !important;
+}
+.gr-textbox label,
+.gr-label label {
+    color: #202123 !important;
+}
+.gr-textbox input[type="text"] {
+    background-color: #f9fbfd !important;
 }
 """
 
-with gr.Blocks(css=css) as iface:
-    gr.Markdown("## تحليل المشاعر بالعربية 📝", elem_id="title")
-    gr.Markdown("أدخل جملة لتحليل المشاعر: هل هي **إيجابية** أم **سلبية**؟", elem_id="description")
-    text_input = gr.Textbox(lines=3, placeholder="اكتب جملتك هنا...", label="النص")
-    predict_btn = gr.Button("تنبؤ")
-    sentiment_output = gr.Label(label="النتيجة")
-    confidence_score = gr.Textbox(label="نسبة الثقة")
+# Build Gradio interface with Blocks to add CSS
+with gr.Blocks(css=custom_css) as iface:
+    gr.Markdown("## تحليل المشاعر العربية بالذكاء الاصطناعي")
+    gr.Markdown("اكتب جملة لتحليل المشاعر (إيجابي أو سلبي)")
+
+    input_text = gr.Textbox(lines=2, placeholder="اكتب الجملة هنا...")
+    sentiment_label = gr.Label(num_top_classes=2, label="المشاعر")
+    confidence_score = gr.Textbox(label="ثقة النموذج")
 
-    def on_predict(text):
-        label, confidence = predict_sentiment(text)
-        confidence_percent = f"{confidence*100:.1f}%"
-        return label, confidence_percent
+    def wrapped_predict(text):
+        return predict_sentiment(text)
 
-    predict_btn.click(fn=on_predict, inputs=text_input, outputs=[sentiment_output, confidence_score])
+    btn = gr.Button("تحليل")
+    btn.click(fn=wrapped_predict, inputs=input_text, outputs=[sentiment_label, confidence_score])
 
 if __name__ == "__main__":
     iface.launch()